Simulations are successfully utilized to reproduce the behavior of complex systems in many knowledge fields. The computational effort is a key factor when high-cost simulations are required in optimization, principally, if the system to be optimized operates under uncertain conditions. In this context, surrogate modeling is useful to alleviate the CPU time. Hence, this paper presents a methodology to assess three surrogate techniques based on genetic programming (GP), a radial basis function neural network (RBF-NNs), and universal Kriging. These techniques are used in this paper to obtain analytical optimization functions that are accurate, fast to evaluate and suitable for interval robust optimization. The experiments were performed in a robust version of the TEAM 22 problem. The results show that the surrogate models obtained are reliable and appropriate for interval robust methods. The methodology presented is flexible and extensible to other problems in diverse fields of interest.